Online |
D0 Grid |
Trigger Simulation |
Graphics |
Level 3 |
Algorithms |
Stu Fuess, 8/28/2002
There weren't any "big" areas of activity in the Online during August, just a number of incremental improvements:
- survived the RunIIb Director's review, prepared for the Sept. Lehman Review
- numerous upgrades made to the electronic logbook and web browser
- continual efforts to clean up startup scripts
- working on system issues, particularly NFS lockups; addressed by moving to 2.4.19 kernel
- control room upgrades ordered
- prepared for another 32 L3 nodes to arrive
- making progress updating Examines
Igor Terekhov, 9/3/2002
Designed "sam call-out"
from Condor MMS to stations and DBS in order to compute in the first approximation the job placement
In grid, worked on the (re-)design of the JIM information model. There was an initial implementation of the monitoring system where SAM concepts are
reflected on the monitoring page that uses HTTP-PHP-LDAP-MDS path. Modified, with the students, the information tree so as to enable browsing
of dynamically created contexts. This will be useful to view the current state of the system. The prototype was attached to a real SAM station,
displaying projects on d0 central-analysis.
Severe performance slow-down from "sam" command-line interfaces forced
us to have an optimized SAM infrmation provider that uses a C++ client to dump SAM servers with subsequent parsing. This made the prototype much
more 'alive'.
Configuration of MDS server and its information providers takes a lot of time - we want to do it correctly so as to incorporate the various use cases
and have easy ways of configuring SAM stations. Some information about SAM stations has been designed to be provided for the Condor ClassAds
for the Job Management service.
We will want to use XML for jim info providers configuration management as well as (possibly) representing some relevant information about
the resources. This design allows us to keep the MDS and Condor branches of the project "relatable" but separate and be
"technology-independent" when informarion is gathered.Oversaw mini-design bythe students of the
back-end information gathering for ClassAds.preferences based on cached data. Condor didn't seem to work when we replaced Script
invocations with compiled functions (for caching and other efficiency).
Oversaw min-doc on design of job submission and specification, by a student.
Continued design sessions with Condor team, tackled the issues of multi-stage proxy forwarding. Set a goal with them to lift the requirement
of running condor-g servers as root by mid-October. This will facilitate
samgrid deployment.
Dugan O'Neil, 9/4/2002
We are currently certifying p12.02.00. All plots are available for users and developers' comment. We hope that p12.03.00 is able to run online in both L2 and L3.
The memory overwrite problem reported last week seemed to be something funny in HepTuple. In any case, it was reported in p11.09.00 but is not reproducible in p11.11.00 or p12. It is no longer high on the priority list.
Improvements have been made to the d0tools interface to the package. It is now possible to use -format=data,mc on the commandline to switch between data and MC modes of running. The -forcercpoutfile option was made default. Also, it will now attempt to automatically generate a list of branches for L3 to put into the ntuple.
Improved default trigger lists have been made part of the d0trigsim package in p12.
We are investigating moving to the bugzilla bugtracking system for problems with d0trigsim. Expect it to be fully functional by
Laurent Duflot, 9/4/2002
I/ d0ve
major changes:
- much improved representation of tracks in the 2D views. Include
elastic_reco to help reconstruct low pT tracks. (A. Haas)
- improved calorimeter lego plot, better looking, add scale lines. Allows to run a special version online that runs only calorimeter
unpacking, so that the display is faster. Useful for calorimeter shifters (T Burnett)
- revive the vertex display (G. Watts)
details can be found at
http://d0.phys.washington.edu/projects/live_event/august_02.htm
TODO:
- investigate using threads to allow for refresh and GUI interaction while events are reconstructed.
- optionally filter events by trigger to show more interesting events
- represent vertices on the 2D views
- display triggers that passes
- define default viewpoints for the control display when the layout of monitors is finalized
- move track code from d0ve_smt to d0ve_reco
II/ d0scan
- release of the XY and RZ views (T. Adams)
- switch to emissive colors to allow vector postscript printing (G. Alverson)
- release of the (eta,phi) flat view (L. Duflot)
- release of the physics lego view (L. Duflot)
- The object dump (printout after selection) from any view now goes to a single window (G. Alverson)
- synchronize with iguana v3.1.1 (GUI+tools) (G. Alverson)
- infrastructure work (iguana, OIV v3) (G. Alverson)
TODO:
- finish the implementation of the XY and RZ views
- L1 Cal trigger tower display for the (eta,phi) view
- L3 jets, electrons and muons for the physics lego view
- display trigger that passes
- solve the font problem under Linux
- certify the code with oiv v3
III/ priorities
a number of items on the priority list established at the Oklahoma workshop
(see July report) were addressed:
- calorimeter only display for d0ve
- XY and RZ (partially), (eta, phi) and physics lego (partially) views released in d0scan
Dan Claes, 9/4/2002
l3fanalyze.....................................................Jon Hayes
A change to ensure shorter variable names has been implemented. To avoid ambiguities, a block name + variable name should now always be specified. Jon's tool list generator will be run for reco farm production, with the list copied to the run_data area. This must be updated every time the farm processes data corresponding to different trigger lists.
L3 Thumbnail.............................................Peter Tamburello
Work has begun defining the TMB information to be stored from the L3Chunk for scripts and filters. At minimum physics object p_t, eta, phi should allow matched to objects at across levels (L1/L2/reco/MC). It was generally agreed that we must store all of quantities L3 selects on, but debated whether we always need to be able to identify which physics_object caused a filter to fire. "Charge" should be added into the PhysResults base class to simplify its appearance in new track objects. It remains unresolved whether the stand-alone tracker is promoted to physics tool or a physics tool wrapper calls it and controls/creates the PhysResults?
Profiling and timing studies.......................................Han Do
Work has begun on timing studies using both stat_manager information and profiling. A modification to d0tools provided by Jon Hays allows running runD0TrigSim with the -profile option. The variety of syntax employed in error messages thwarted an initial plan to retrieve some timing info by parsing log files. Additionally it appears that not all tools correctly implement calls to l3StatDevel. With timestamps in log files neither reliable nor precise enough for our needs, the focus will be on the monitor statistics with profiler output as a consistency check. Initial profiler results on L3 tool times based on 1000 recent data events have been shown. The next step will make a more detailed breakdown of where the time is being used for the 10 most time consuming tools.
Single electron triggers...............................Ulla Blumenschein
1720 L1 CEM(1,20) Mark&Passed events (global_CMT-8.10) were used to check the rejection rates for the new set of 5 single electron filter scripts. These values confirm the online reports, while the .or. of the 5 filter scripts is 14+/-1. An anti-correlation between events passing the calorimeter-based filters and those passing the track-based filter for CEM(1,10) suggests the former include many high energy pizeros (no corresponding high p_t tracks). This needs to be verified through MC study. Plots from the special Run 150490 showed the pt spectrum for the leading L3Electron clusters from Marked&Passed EM_HI triggers: with no cuts, EM fraction > 0.9 and EM fraction + shower width cuts. The L3 pass rate (with respect to L1) was presented as a function of a Pt min cut. At 40 GeV this was around 3.7 % dropping down to 0.7% at 50 Gev.
L3 Tracking..............................................Daniel Whiteson
A study of L3 track-matched electrons in a Z->ee data sample was made to measure the global tracker's effficiency. These events were selected through calorimeter information (and a di-EM mass cut), unbiased by any track information. Matching was made to the nearest track in eta, phi. Correcting for track curvature (and achieving better propagation to the calorimeter) showed overall efficiciencies run 50-60% (70% in the later runs and 70-80% in the central eta region). Event details of the inefficiencies have not yet been looked at. The discrepancy reported earlier between online and predicted rejections are due to the addition
of L2mu(5). The original prediction of 56 was wrt to L1. The online rejection of 15 is wrt L2. With an advertised L2mu(5) rejection rate of about 5, these values seem consistent. Cookbook-level documentation on running L3 tracking on data and Monte Carlo is now available at
http://www-d0.fnal.gov/~danielw/tracking/L3TrackingFAQ.html.
Muon certification........................Martin Wegner,Martijn Mulders
With global_CMT-8.00 Robert Illingworth's geometry manager has made the offline geometry available in L3. Other recent changes (e.g., survey geometry) bring L3 and offline Pt resolutions in line, but L2 discrepancies still not understood. For the next Trigger Board presentation rejection and efficiencies will be shown with respect to L2.
L3MEt update..................................................Lee Sawyer
A first look at real data showed MEt distributions (x,y and totals) has compared before/after the offline zero suppression change. This work does not yet utilize the L3NADA information, and still uses the null vertex. Phi distributions show the vertexing offset. Studies continue to check if Et distributions can be cleaned up by the current CFT z vertex, bad calorimter run list, non-linearity corrections. Zero-suppression is available off-line, but should Marumi include the ped subtraction online? The initial aim is to add a missing E_T filter script (possibly plus a loose calorimeter electron) to provide extra redundancy for W->enu events.
Harry Melanson, 9/5/2002
Status of releases
Recent progress:
Major outstanding problems:
Ongoing projects: